# Anthropic Introduces **Claude Skills** — Modular Capabilities for AI Agents

This morning, Anthropic announced [Claude Skills](https://www.anthropic.com/news/skills), a **new approach to adding specialized capabilities** to their models.

> Claude can now use *Skills* to improve performance on specific tasks. Skills are folders containing instructions, scripts, and resources that Claude can load when needed.  
> Skills are activated only when relevant, enabling Claude to handle specialized work such as Excel processing or applying brand guidelines.

📚 See details in:
- [Anthropic engineering blog](https://www.anthropic.com/engineering/equipping-agents-for-the-real-world-with-agent-skills)
- GitHub repo: [anthropic/skills](https://github.com/anthropics/skills)

💡 I accidentally preempted this announcement after [reverse-engineering it last Friday](https://simonwillison.net/2025/Oct/10/claude-skills/).

---

## What Are Skills?

**Concept**:  
A skill is **a Markdown file** that describes how the model should perform a specific task.  
Optional: extra **documents** and **scripts** to accelerate or make the process more reliable.

Recent features like [Claude’s document creation tools](https://www.anthropic.com/news/create-files) — launched alongside [the code interpreter](https://simonwillison.net/2025/Sep/9/claude-code-interpreter/) — are implemented entirely as skills.

📂 Available skills include creating `.pdf`, `.docx`, `.xlsx`, and `.pptx` files:  
[Document skills](https://github.com/anthropics/skills/tree/main/document-skills) on GitHub.

---

## Token Efficiency

At session start, Claude’s harness scans available skill files and reads their short YAML **frontmatter descriptions**.  
Each skill adds only **a few dozen tokens**, loading details only when relevant — maximizing efficiency.

---

## Related: AI Content Monetization

For creators building workflows around Claude Skills, platforms like [AiToEarn官网](https://aitoearn.ai/) can help:

- **Open-source** AI content monetization platform
- Unified creation + publishing + analytics  
- Supports Douyin, Kwai, WeChat, Bilibili, Xiaohongshu, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, X (Twitter)
- Includes AI model ranking: [AI模型排名](https://rank.aitoearn.ai)

This enables smooth deployment of skill-generated outputs to multi-platform audiences.

---

## Navigation

- [Trying out the slack-gif-creator skill](#trying-out-the-slack-gif-creator-skill)
- [Skills depend on a coding environment](#skills-depend-on-a-coding-environment)
- [Claude Code as a General Agent](#claude-code-as-a-general-agent)
- [Skills compared to MCP](#skills-compared-to-mcp)
- [Here come the Skills](#here-come-the-skills)
- [The simplicity is the point](#the-simplicity-is-the-point)

---

## Trying Out the `slack-gif-creator` Skill

**Metadata excerpt** for [slack-gif-creator skill](https://github.com/anthropics/skills/blob/main/slack-gif-creator/SKILL.md):

> Toolkit for creating animated GIFs optimized for Slack, with validators for size constraints and composable animation primitives.

---

### Why It Matters
- Lightweight metadata + modular skills → **lower token use**
- Instant discoverability and contextual activation
- Complements publishing workflows (via platforms like [AiToEarn官网](https://aitoearn.ai/))

---

### Test Example

Steps:
1. Enable **slack-gif-creator** in Claude's [settings](https://claude.ai/settings/capabilities).
2. Prompt:  
   > Make me a gif for Slack about how Skills are way cooler than MCPs
3. Result: [GIF generated](https://claude.ai/share/eff7ae7b-b386-417b-9fa0-213fa76ace6e) *(click-to-play due to flashing)*

![image](https://blog.aitoearn.ai/content/images/2025/10/img_001-5.gif)

---

### Python Script Excerpt

import sys

sys.path.insert(0, '/mnt/skills/examples/slack-gif-creator')

from PIL import Image, ImageDraw, ImageFont

from core.gif_builder import GIFBuilder

... GIF build code ...

info = builder.save('/mnt/user-data/outputs/skills_vs_mcps.gif',

num_colors=128,

optimize_for_emoji=False)

print(f"GIF created successfully!")

print(f"Size: {info['size_kb']:.1f} KB")

print(f"Frames: {info['frame_count']}")

print(f"Duration: {info['duration_seconds']:.1f}s")

**Slack file size check:**

passes, check_info = check_slack_size('/mnt/user-data/outputs/skills_vs_mcps.gif', is_emoji=False)

if passes:

print("✓ Ready for Slack!")

else:

print(f"⚠ File size: {check_info['size_kb']:.1f} KB (limit: {check_info['limit_kb']} KB)")

---

## Skills Depend on a Coding Environment

**Requirement**:  
- Filesystem access  
- Navigation tools  
- Command execution capability

**Why**:  
- Same pattern as ChatGPT Code Interpreter (2023) and tools like Cursor, Claude Code, Codex CLI, Gemini CLI.
- Key difference: dependency on execution environment vs heavier protocols like MCP.
- **Safe sandboxing** is critical to mitigate risks (e.g., prompt injection attacks).

---

## Claude Code as a General Agent

Prediction fail: 2025 → **Year of agents**.  
[Claude Code](https://www.claude.com/product/claude-code) enables general automation — not just coding.

**Example Data Journalism Agent Build:**
- Skills for US Census data handling  
- Skills for SQLite/DuckDB integration  
- Skills for publishing datasets online  
- Skills for D3 visualizations and story discovery

---

## Skills Compared to MCP

[MCP](https://modelcontextprotocol.io/) → high token overhead  
- GitHub MCP: tens of thousands of tokens consumed before the model can work  
- CLI tools often achieve the same goals with far fewer tokens

**Skills advantage**:  
- Drop in a Markdown file — no new CLI tool needed.

---

## Here Come the Skills

Sharing ease:
- Many can be a single file  
- Larger, more complex = folder with multiple files

Resources:
- [Agent Skills documentation](https://docs.claude.com/en/docs/agents-and-tools/agent-skills/overview)
- [Claude Skills Cookbook](https://github.com/anthropics/claude-cookbooks/tree/main/skills)

---

## Model-Agnostic Design

Skills can be run by:
- Codex CLI  
- Gemini CLI  
…or any tool that can read Markdown and execute scripts.

---

## A Coming Cambrian Explosion

Expect rapid adoption — easier than MCP, broadly compatible.

---

## The Simplicity Is the Point

Critics: Skills are "too simple."  
Reality: **Simplicity = power**.

- MCP: complex protocol  
- Skills: Markdown + optional scripts

LLM harness manages the complexity → robust, efficient workflows.

---

**Extra note for creators**:  
Platforms like [AiToEarn官网](https://aitoearn.ai/) align perfectly with Skills’ lightweight nature — connecting AI-generated outputs to monetization across global channels quickly and simply.